GaN Laser Diodes

Physics and Simulation of Optoelectronic Devices VIII, ed. R. Binder, P. Blood & M. Osinski, SPIE Proc. 3944, 2000.

Simulation and Optimization of 420 nm InGaN/GaN Laser Diodes Joachim Piprek* a, R. Kehl Sink a, Monica A. Hansen b, John E. Bowers a, and Steve P. DenBaars b a

Electrical and Computer Engineering Department b Materials Department University of California at Santa Barbara, Santa Barbara, CA 93106 ABSTRACT Using self-consistent laser simulation, we analyze the performance of nitride Fabry-Perot laser diodes grown on sapphire. The active region contains three 4 nm InGaN quantum wells. It is sandwiched between GaN separate confinement layers and superlattice AlGaN/GaN cladding layers. AlGaN is used as an electron barrier layer. Pulsed lasing is measured near 420 nm wavelength and at temperatures up to 120o C. Advanced laser simulation is applied to link microscopic device physics to measurable device performance. Our two-dimensional laser model considers carrier drift and diffusion including thermionic emission at hetero-boundaries. The local optical gain is calculated from the wurtzite band structure employing a nonLorentzian line broadening model. All material parameters used in the model are evaluated based on recent literature values as well as our own experimental data. Simulation results are in good agreement with measurements. Multi-lateral mode lasing is calculated with a high order vertical mode. The carrier distribution among quantum wells is found to be strongly non-uniform leading to a parasitic (absorbing) quantum well. The influence of defect recombination, vertical carrier leakage and lateral current spreading is investigated. The reduction of such carrier losses is important to achieve lower threshold currents and less self-heating. Several device optimization options are proposed. Elimination of the parasitic quantum well is shown to substantially enhance the device performance. Keywords: InGaN, blue laser diode, nitride semiconductor laser, quantum well devices, numerical analysis, semiconductor device modeling

1. INTRODUCTION Violet InGaN/GaN laser diodes have been pioneered by Nakamura et al.1 and are now commercially available.2 These laser diodes have potential in a number of applications such as optical storage, printing, full-color displays, chemical sensors and medical applications. The data capacity of digital versatile disks (DVDs) can be increased from 4.7 to more than 15 gigabites by using InGaN lasers instead of red AlInGaP lasers.1 The estimated InGaN laser lifetime under room-temperature (RT) continuous-wave (CW) operation has been improved to more than 10,000 hours by using laterally epitaxially overgrown GaN as substrate and AlGaN/GaN strained-layer superlattices (SLS) as cladding layers.3 Threshold current densities as low as 1.2 kA/cm2 were achieved using In 0.1 5 Ga0.85N quantum wells (QWs).4 Fundamental transverse mode operation up to 30 mW output power and with a slope efficiency as high as 1 W/A was obtained using a ridge-waveguide laser with 2 µm ridge width.5 These lasers have cleaved facets with one facet coated by two pairs of quarter-wave TiO2 /SiO2 dielectric multilayers. Since the first announcement of a nitride laser diode by Nakamura et al. in December 1995, more than 10 other research groups have succeeded in fabricating similar laser diodes. However, only about half of these groups have achieved RT-CW lasing so far.6 In 1997, researchers of our university demonstrated an InGaN laser diode employing uncoated reactive ion etched (RIE) facets.7 The structural properties of RIE facets are far from ideal leading to poor optical feedback. Despite several design improvements during the last two years, RT-CW operation has not yet been achieved with these RIE lasers.8 Major technological challenges remain as well as the need for a more detailed understanding of nitride laser physics. Advanced numerical laser simulation can help to establish quantitative links between microscopic material properties and measured device performance. *

Correspondence: Email: [email protected], WWW: http://eci.ucsb.edu/~piprek, Telephone: 805-893-4051

In this paper, we present first results of a self-consistent numerical simulation of our InGaN laser diodes. We use a commercial laser diode software package9 that combines band structure and gain calculations with two-dimensional (2D) simulations of wave guiding, current flow, and heat flux. The model considers carrier drift and diffusion including thermionic emission at hetero-boundaries. This allows for a study of vertical carrier leakage, lateral current spreading, and defect recombination. The reduction of such carrier losses is important to achieve lower threshold currents and less selfheating. The local optical gain is calculated from the wurtzite band structure including valence mixing and strain effects. A non-Lorentzian line broadening model is utilized and many-body effects are considered. Transversal multimode lasing is taken into account including higher order vertical modes. Such a comprehensive model requires more than 40 material parameters per epitaxial layer, most of which are not exactly known. We carefully evaluated all material parameters based on recent literature values as well as our own experimental data. However, material parameters remain the main source of uncertainty in our model. Section 2 gives a brief description of device structure and experimental results. Section 3 provides details of the theoretical model and of material parameters used. Section 4 presents a microscopic analysis of laser physics based on simulation results. Section 5 discusses device optimization options.

40

Light Output (mW)

35 LEO

30 25 20 15 10

planar

5 0 0

500

1000

1500

2000

2500

3000

Current (mA)

Fig. 1: Schematic cross section of our laser diodes.

Fig. 2: Measured light vs. current characteristics.

2. DEVICE STRUCTURE AND EXPERIMENTAL RESULTS The schematic structure of our InGaN multiple-quantum-well (MQW) laser diodes is given in Fig. 1. The laser is grown on a lateral epitaxial overgrowth (LEO) GaN substrate to reduce the dislocation density. Details of growth and processing are described elsewhere.10 Table 1 lists compositions, layer thicknesses, and doping densities. To achieve sufficient waveguiding, the Al0.09Ga 0.91N cladding regions need to be relatively thick and they are grown as Al0.18Ga 0.82N/GaN superlattices. Laser facets are formed by Cl2 reactive ion etching (RIE) of 45 µm wide mesas of various lengths ranging from L = 400 µm to 1600 µm. Contact stripes are patterned on these mesas with widths ranging from W = 5 µm to 15 µm. The structure was etched around the p-contact stripe through the p-cladding for index guiding. The n- and p-contacts are formed by electron beam evaporation of Ti/Al and Pd/Au, respectively. Electrical testing was performed using 50 ns pulses with a 1 kHz pulse repetition rate. Figure 2 shows the typical pulsed light output per uncoated facet of a laser diode grown on LEO GaN and a laser diode grown on planar GaN as a function of forward current. Due to a reduction in nonradiative recombination, the minimum threshold current density jth is reduced from 10 kA/cm2 for laser diodes grown on planar GaN to 4.8 kA/cm2 for laser diodes grown on LEO GaN. The latter exhibit an improved slope efficiency ηd up to 3.2 % per facet. However, there are large variations among devices and typical numbers for LEO GaN lasers as shown in Fig. 2 are jth = 12 kA/cm2 and ηd = 0.04 W/A = 1.3% (W=5 µm, L=1600 µm). The measured slope efficiency does not represent the entire light

2

emitted on each side. About 60 % of the light power is not detected, which is attributed to photon scattering at the rough RIE facet, to its tilted angle, and to laser beam interference with the remaining GaN substrate.11 The facet reflectance is estimated to be only R = 0.05, considerably less than the ideal value of 0.18.12 Considering these numbers, the inverse slope efficiency vs. cavity length plot gives ηi = 22 % internal efficiency and αi = 42 cm-1 internal optical loss for our LEO GaN lasers.10

LAYER contact cladding waveguide electron stopper waveguide barrier quantum well barrier quantum well barrier quantum well barrier waveguide cladding compliance layer substrate

MATERIAL GaN Al0.18Ga 0.82N/GaN SL GaN Al0.2 Ga 0.8 N GaN In 0.04Ga 0.96N In 0.1 3 Ga 0.87N In 0.04Ga 0.96N In 0.13Ga 0.87N In 0.04Ga 0.96N In 0.13Ga 0.87N In 0.04Ga 0.96N GaN Al0.18Ga 0.82N/GaN SL In 0.05Ga 0.95N GaN

THICKNESS [nm] 100 451 90 20 8.5 8.5 4.0 8.5 4.0 8.5 4.0 8.5 98.5 451 100 2000

DOPING [1018 cm-3 ] 75 (Mg) 75 (Mg) 75 (Mg) 100 (Mg) 0.05 (Si) 0.05 (Si) 0.05 (Si) 6.5 (Si) 0.05 (Si) 6.5 (Si) 0.05 (Si) 6.5 (Si) 6.5 (Si) 2.0 (Si) 6.0 (Si) 6.0 (Si)

Table 1: Layer parameters as used in the simulation.

3. THEORETICAL MODEL AND MATERIAL PARAMETERS LASTIP self-consistently combines 2D carrier transport, heat flux, optical gain computation, and wave guiding within the transversal plane (x,y). Longitudinal variations are of minor importance in our device. Heat flux calculations are not included in our analysis of pulsed laser operation. Further details of the laser model are published elsewhere.13 We discuss here only those aspects that are crucial to our analysis. The drift-diffusion model of carrier transport includes Fermi statistics and thermionic emission at hetero-barriers. Thermionic emission is mainly controlled by the offset of conduction band (∆Ec ) and valence band (∆Ev ) at hetero-barriers. For AlGaN/GaN and InGaN/GaN we use a band offset ratio (∆Ec/∆Ev ) of 0.67/0.33 14 and 0.3/0.7,15 respectively. Due to the small conduction band offset in the InGaN MQW, vertical electron leakage into the p-side GaN waveguide layer is a crucial issue with nitride lasers, requiring an AlGaN barrier layer above the MQW (cf. Fig. 1). The room temperature band gap Eg of our ternary alloys is calculated from 14, 16

Eg (AlxGa 1-xN) [eV] = 6.28 x + 3.42 (1-x) - 0.98 x (1-x)

(1)

Eg (In xGa 1-xN) [eV] = 1.89 x + 3.42 (1-x) - 3.8 x (1-x)

(2)

A major handicap with nitride semiconductors is the high activation energy of the Mg acceptor which causes the hole density to be considerably smaller than the Mg density. We use an Mg activation energy of 170 meV for GaN 17 which is assumed to increase by 3meV per % Al for AlGaN.6 A uniform hole mobility of 2 cm2 /Vs is assumed 18 which is only a

3

rough estimate, in particular for the p-side AlGaN/GaN superlattice. In n-type material, the Si donor activation energy is about 20 meV. 19 Our electron mobility values are 200 cm2 /Vs for GaN, 100 cm2 /Vs for InGaN, and 30 cm2 /Vs for AlGaN.18 Within passive layers, a spontaneous emission parameter of B = 2 × 10-10 cm3 s -1 is employed.20 The spontaneous recombination rate in quantum wells is much larger than in passive layers and it is calculated self-consistently by integration of the spontaneous emission spectrum. The Shockley-Read-Hall (SRH) recombination lifetime of electrons and holes is assumed to be 1 ns, however, this is only a rough estimate since type and density of recombination centers are very much technology dependent. The GaN Auger parameter C = 2 × 10-31 cm6 /s is relatively small,21 nevertheless, Auger recombination might become important at high carrier densities.

0.00

6000 HH1

-0.05

LH2 HH3

-0.10

Gain [1/cm]

Valence Band Energy [eV]

LH1 HH2

LH3 CH1 HH4

-0.15

19

-3

5000

N [10 c m ] = 6

4000

5

3000

4

2000 3

LH4 1000

-0.20 0.00

0.05

0.10

0.15

0.20

In-Plane Wave Number k x=k y [1/A]

Fig. 3: In-plane dispersion of valence subbands.

0 0.39

0.40

0.41

0.42

Wavelength [ µm ]

Fig. 4: Gain spectra for different carrier densities N.

In our strained In 0.13Ga 0.87N/In 0.04Ga 0.96N quantum wells, the conduction bands are assumed to be parabolic and the nonparabolic valence bands are computed by the kp method including valence band mixing.22 Valence band effective mass parameters (A 1 - A 6 ) are obtained from Ref.23 , GaN values are used for the crystal field split energy ∆1 = 16 meV and for the spin-orbit split energies ∆2 = ∆3 = 4 meV as well as for the deformation potentials.22 Elastic constants are linearly interpolated between C13 = 114 GPa and C33 = 381 GPa given for GaN and C13 = 94 GPa and C33 = 200 GPa given for InN.24 Vegard's law is applied to obtain the ternary lattice constant from a = 0.3548 nm (InN), 0.3189 nm (GaN), and 0.3112 nm (AlN).25 These parameters give 1.4 % compressive biaxial strain in our quantum wells and 0.4 % in the barriers. The calculated heavy-hole (HH), light-hole (LH) and crystal-field split-hole (CH) bands are shown in Fig. 3, illustrating the strong valence band mixing. Similar band structures of InGaN quantum wells have been calculated by Yeo et al.26 For structures like ours, grown along the c-axis, the quantum well strain is biaxial, lowering the CH band but hardly separating HH and LH bands, i.e., strain is not as effective in reducing the threshold carrier density as in GaAs or InP based lasers. Uniaxial strain promises lower threshold currents.27 For gain computation, a non-Lorentzian line broadening model is utilized following Asada 28 and using a longitudinal optical phonon energy of 90 meV. 25 At room temperature and with N = 1019 cm-3 carrier density, the effective intraband relaxation time becomes 15 fs which is in good agreement with other publications.29, 30 Band gap shrinkage caused by carrier-carrier interaction is considered as ∆Eg = - ξ N1/3 , the renormalization parameter ξ = 4.5 × 10-8 eVcm for our 4 nm thick quantum wells is extracted from Ref.31 . Recently, there have been several publications on piezoelectric field effects on spontaneous and stimulated emission in InGaN quantum wells (see, e.g., Ref.30 and Ref.32 ). The wider the quantum well, the stronger electrons and holes are separated by the piezoelectric field, thereby reducing optical gain as well as spontaneous emission. However, screening by quantum confined carriers 30 and mainly by free carriers 33 is expected to suppress piezoelectric field effects at typical threshold carrier densities of 3 ×1019 cm-3 . The high threshold carrier density is also assumed to eliminate exciton enhancements of the gain, despite the large exciton binding energy in nitrides.34 Considering a GaN exciton Bohr radius of rB =1.6 nm, 31 the excitonic phase becomes unstable as the 2D electron-hole density exceeds 1/πrB2 = 1.2×1013 cm-2 ,

4

corresponding to N = 3 × 1019 cm-3 in our quantum wells. Structural non-uniformities within the quantum well may also affect the optical gain.35 However, QW carrier localization does not seem to dominate the gain of nitride lasers.1 In summary, the gain mechanism in real nitride lasers is a very complex issue and it is not yet fully understood. Comprehensive laser simulation will help to link microscopic gain mechanisms to the macroscopic device performance. Assuming uniform rectangular quantum wells, we obtain for our laser the gain spectra given in Fig. 4. The gain peak near 409 nm includes contributions from the HH1 subband as well as from the LH1 subband (cf. Fig. 3).

2.0

Refractive Index

2.5

1.5 st

1 vertical mode

2.4

th

lasing mode (5 ) 1.0

2.3

2.2

0.5

2.1

Mode Intensity (normalized)

2.6

0.0 2.0 0.0

0.5

1.0

1.5

2.0

2.5

3.0

Vertical Axis [ µm] Fig. 5: Vertical profiles of refractive index and optical modes.

Our simulation of the 2D optical waveguide adopts the measured parameters for the internal loss (αi = 42 cm-1 ) and for the power reflectance per facet (R = 0.05) giving a distributed mirror loss coefficient of αm = 19 cm-1 for the longest cavity length (L = 1600 µm) which will be used in the following. The refractive index of ternary alloys is extracted from GaN waveguide measurements 36 using bandgap variations in the model (x